Displays overlapping token logos for baskets, pools, routes, and portfolio rows.
Displays overlapping token logos for baskets, pools, routes, and portfolio rows.
Text arriving token by token with a caret, like a model's reply; or follow a real stream.
LLM token usage meter showing tokens used against a monthly limit, a labelled progress bar and a percent of budget note with the reset date. Useful in AI apps.
Token usage visualization with input/output/total counts, cost estimation for popular models, and interactive token breakdown with hover-to-inspect boundaries.
The Toman currency symbol, as a standalone icon component.
A filled secondary button in two quiet fills: a tinted tonal fill and a neutral secondary one, as in Steeze's Earnings and Edit Profile pair.
Area chart with rotated SVG screentone halftone bands, speech-balloon peak callout, keyboard navigation, and accessible data table.
Quill's uppercase tag pill for status, tiers, and labels — wraps Badge with the pigment vocabulary (moss/gold/terracotta/indigo/neutral/muted) baked in so tone never drifts per call site.
Interactive ranked list with screentone progress bars, share percentages, and impact-burst leader badge.
方条左侧标签、右侧一排色点的单选组,色点取主题变量,跟随亮暗与颜色主题。
Compact daily bar sparkline filled with 45-degree SVG screentone dots.
A read-only tracking-quality gauge built from real pivoted-tonearm geometry: a headshell sweeps a 215mm-pivot arm across a spinning record from lead-in to run-out while a small error needle reads the actual geometric tracking-error function, centering exactly at the two Baerwald null radii (66mm, 120mm) and drifting off zero everywhere else.
A loader / progress surface that prints its content in the way a laser printer images a page: a top-to-bottom expose/develop wipe with toner visibly catching behind it, fine edges thinning for 400ms before filling to full density, and a persistent starvation streak on any large fill area, locked down by a single fuse-roller highlight pass. An unbounded charge/wipe/hold/reset loop, not a spinner.
A tool call: name, live status, timing, parameters and result.
Tool call component with state-aware header and JSON input/output blocks
Displays tool call details including input, output, status, and errors. Ideal for visualizing AI tool usage in chat UIs.
Quiet collapsible tool call with input and output.
The tools an AI assistant used to get its answer, step by step: their status, how long they took, and their input and output.
Asks before an agent runs a tool. Shows what the call will do, offers Allow once, Always allow and Deny, and leaves a one-line record of the answer.
Approve or deny an AI SDK tool call, or a batch of them, with a gesture that matches its risk: undo for low, click again for medium, hold for high and a typed phrase for critical. Pass addToolApprovalResponse from useChat as onRespond.
Your product between the tools it connects, with data flowing along curved beams; hover to focus, click to pause.
Display AI tool/function invocations with status indicators, input/output visualization, and collapsible interface for MCP and function calling workflows.
One tool invocation in a disclosure card: status dot, tool name, duration, and the request and result payloads pretty-printed on expand
What an agent did, one row per call: a status mark (a still half circle while it runs, never a spinner), the action in words, the target in mono cut at the end, what it found ("5 sources", "+2 −2") and the time. A call with an input or an output opens them under it, in a mono well; a part with its own surface (a log, a diff, a list of sources) opens as given in details. A group folds a run of calls and closes itself when they finish, and an approval asks first on a raised panel with Allow as the accent fill and Deny beside it, then becomes a row that says what was chosen.
Showing 28,657–28,680 of 29,980
Showing 28,657–28,680 of 29,980